xref: /netbsd-src/sys/kern/vfs_mount.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: vfs_mount.c,v 1.70 2019/02/20 10:08:37 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997-2011 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: vfs_mount.c,v 1.70 2019/02/20 10:08:37 hannken Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/kernel.h>
74 
75 #include <sys/atomic.h>
76 #include <sys/buf.h>
77 #include <sys/conf.h>
78 #include <sys/fcntl.h>
79 #include <sys/filedesc.h>
80 #include <sys/device.h>
81 #include <sys/kauth.h>
82 #include <sys/kmem.h>
83 #include <sys/module.h>
84 #include <sys/mount.h>
85 #include <sys/fstrans.h>
86 #include <sys/namei.h>
87 #include <sys/extattr.h>
88 #include <sys/syscallargs.h>
89 #include <sys/sysctl.h>
90 #include <sys/systm.h>
91 #include <sys/vfs_syscalls.h>
92 #include <sys/vnode_impl.h>
93 
94 #include <miscfs/genfs/genfs.h>
95 #include <miscfs/specfs/specdev.h>
96 
97 enum mountlist_type {
98 	ME_MOUNT,
99 	ME_MARKER
100 };
101 struct mountlist_entry {
102 	TAILQ_ENTRY(mountlist_entry) me_list;	/* Mount list. */
103 	struct mount *me_mount;			/* Actual mount if ME_MOUNT,
104 						   current mount else. */
105 	enum mountlist_type me_type;		/* Mount or marker. */
106 };
107 struct mount_iterator {
108 	struct mountlist_entry mi_entry;
109 };
110 
111 static struct vnode *vfs_vnode_iterator_next1(struct vnode_iterator *,
112     bool (*)(void *, struct vnode *), void *, bool);
113 
114 /* Root filesystem. */
115 vnode_t *			rootvnode;
116 
117 /* Mounted filesystem list. */
118 static TAILQ_HEAD(mountlist, mountlist_entry) mountlist;
119 static kmutex_t			mountlist_lock;
120 int vnode_offset_next_by_lru	/* XXX: ugly hack for pstat.c */
121     = offsetof(vnode_impl_t, vi_lrulist.tqe_next);
122 
123 kmutex_t			mntvnode_lock;
124 kmutex_t			vfs_list_lock;
125 
126 static specificdata_domain_t	mount_specificdata_domain;
127 static kmutex_t			mntid_lock;
128 
129 static kmutex_t			mountgen_lock;
130 static uint64_t			mountgen;
131 
132 void
133 vfs_mount_sysinit(void)
134 {
135 
136 	TAILQ_INIT(&mountlist);
137 	mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE);
138 	mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE);
139 	mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE);
140 
141 	mount_specificdata_domain = specificdata_domain_create();
142 	mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE);
143 	mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE);
144 	mountgen = 0;
145 }
146 
147 struct mount *
148 vfs_mountalloc(struct vfsops *vfsops, vnode_t *vp)
149 {
150 	struct mount *mp;
151 	int error __diagused;
152 
153 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
154 	mp->mnt_op = vfsops;
155 	mp->mnt_refcnt = 1;
156 	TAILQ_INIT(&mp->mnt_vnodelist);
157 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
158 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
159 	mp->mnt_vnodecovered = vp;
160 	mount_initspecific(mp);
161 
162 	error = fstrans_mount(mp);
163 	KASSERT(error == 0);
164 
165 	mutex_enter(&mountgen_lock);
166 	mp->mnt_gen = mountgen++;
167 	mutex_exit(&mountgen_lock);
168 
169 	return mp;
170 }
171 
172 /*
173  * vfs_rootmountalloc: lookup a filesystem type, and if found allocate and
174  * initialize a mount structure for it.
175  *
176  * Devname is usually updated by mount(8) after booting.
177  */
178 int
179 vfs_rootmountalloc(const char *fstypename, const char *devname,
180     struct mount **mpp)
181 {
182 	struct vfsops *vfsp = NULL;
183 	struct mount *mp;
184 	int error __diagused;
185 
186 	mutex_enter(&vfs_list_lock);
187 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
188 		if (!strncmp(vfsp->vfs_name, fstypename,
189 		    sizeof(mp->mnt_stat.f_fstypename)))
190 			break;
191 	if (vfsp == NULL) {
192 		mutex_exit(&vfs_list_lock);
193 		return (ENODEV);
194 	}
195 	vfsp->vfs_refcount++;
196 	mutex_exit(&vfs_list_lock);
197 
198 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
199 		return ENOMEM;
200 	error = vfs_busy(mp);
201 	KASSERT(error == 0);
202 	mp->mnt_flag = MNT_RDONLY;
203 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
204 	    sizeof(mp->mnt_stat.f_fstypename));
205 	mp->mnt_stat.f_mntonname[0] = '/';
206 	mp->mnt_stat.f_mntonname[1] = '\0';
207 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
208 	    '\0';
209 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
210 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
211 	*mpp = mp;
212 	return 0;
213 }
214 
215 /*
216  * vfs_getnewfsid: get a new unique fsid.
217  */
218 void
219 vfs_getnewfsid(struct mount *mp)
220 {
221 	static u_short xxxfs_mntid;
222 	fsid_t tfsid;
223 	int mtype;
224 
225 	mutex_enter(&mntid_lock);
226 	mtype = makefstype(mp->mnt_op->vfs_name);
227 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
228 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
229 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
230 	if (xxxfs_mntid == 0)
231 		++xxxfs_mntid;
232 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
233 	tfsid.__fsid_val[1] = mtype;
234 	while (vfs_getvfs(&tfsid)) {
235 		tfsid.__fsid_val[0]++;
236 		xxxfs_mntid++;
237 	}
238 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
239 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
240 	mutex_exit(&mntid_lock);
241 }
242 
243 /*
244  * Lookup a mount point by filesystem identifier.
245  *
246  * XXX Needs to add a reference to the mount point.
247  */
248 struct mount *
249 vfs_getvfs(fsid_t *fsid)
250 {
251 	mount_iterator_t *iter;
252 	struct mount *mp;
253 
254 	mountlist_iterator_init(&iter);
255 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
256 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
257 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
258 			mountlist_iterator_destroy(iter);
259 			return mp;
260 		}
261 	}
262 	mountlist_iterator_destroy(iter);
263 	return NULL;
264 }
265 
266 /*
267  * Take a reference to a mount structure.
268  */
269 void
270 vfs_ref(struct mount *mp)
271 {
272 
273 	KASSERT(mp->mnt_refcnt > 0 || mutex_owned(&mountlist_lock));
274 
275 	atomic_inc_uint(&mp->mnt_refcnt);
276 }
277 
278 /*
279  * Drop a reference to a mount structure, freeing if the last reference.
280  */
281 void
282 vfs_rele(struct mount *mp)
283 {
284 
285 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
286 		return;
287 	}
288 
289 	/*
290 	 * Nothing else has visibility of the mount: we can now
291 	 * free the data structures.
292 	 */
293 	KASSERT(mp->mnt_refcnt == 0);
294 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
295 	mutex_destroy(&mp->mnt_updating);
296 	mutex_destroy(&mp->mnt_renamelock);
297 	if (mp->mnt_op != NULL) {
298 		vfs_delref(mp->mnt_op);
299 	}
300 	fstrans_unmount(mp);
301 	/*
302 	 * Final free of mp gets done from fstrans_mount_dtor().
303 	 *
304 	 * Prevents this memory to be reused as a mount before
305 	 * fstrans releases all references to it.
306 	 */
307 }
308 
309 /*
310  * Mark a mount point as busy, and gain a new reference to it.  Used to
311  * prevent the file system from being unmounted during critical sections.
312  *
313  * vfs_busy can be called multiple times and by multiple threads
314  * and must be accompanied by the same number of vfs_unbusy calls.
315  *
316  * => The caller must hold a pre-existing reference to the mount.
317  * => Will fail if the file system is being unmounted, or is unmounted.
318  */
319 static inline int
320 _vfs_busy(struct mount *mp, bool wait)
321 {
322 
323 	KASSERT(mp->mnt_refcnt > 0);
324 
325 	if (wait) {
326 		fstrans_start(mp);
327 	} else {
328 		if (fstrans_start_nowait(mp))
329 			return EBUSY;
330 	}
331 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
332 		fstrans_done(mp);
333 		return ENOENT;
334 	}
335 	vfs_ref(mp);
336 	return 0;
337 }
338 
339 int
340 vfs_busy(struct mount *mp)
341 {
342 
343 	return _vfs_busy(mp, true);
344 }
345 
346 int
347 vfs_trybusy(struct mount *mp)
348 {
349 
350 	return _vfs_busy(mp, false);
351 }
352 
353 /*
354  * Unbusy a busy filesystem.
355  *
356  * Every successful vfs_busy() call must be undone by a vfs_unbusy() call.
357  */
358 void
359 vfs_unbusy(struct mount *mp)
360 {
361 
362 	KASSERT(mp->mnt_refcnt > 0);
363 
364 	fstrans_done(mp);
365 	vfs_rele(mp);
366 }
367 
368 struct vnode_iterator {
369 	vnode_impl_t vi_vnode;
370 };
371 
372 void
373 vfs_vnode_iterator_init(struct mount *mp, struct vnode_iterator **vnip)
374 {
375 	vnode_t *vp;
376 	vnode_impl_t *vip;
377 
378 	vp = vnalloc_marker(mp);
379 	vip = VNODE_TO_VIMPL(vp);
380 
381 	mutex_enter(&mntvnode_lock);
382 	TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vip, vi_mntvnodes);
383 	vp->v_usecount = 1;
384 	mutex_exit(&mntvnode_lock);
385 
386 	*vnip = (struct vnode_iterator *)vip;
387 }
388 
389 void
390 vfs_vnode_iterator_destroy(struct vnode_iterator *vni)
391 {
392 	vnode_impl_t *mvip = &vni->vi_vnode;
393 	vnode_t *mvp = VIMPL_TO_VNODE(mvip);
394 
395 	mutex_enter(&mntvnode_lock);
396 	KASSERT(vnis_marker(mvp));
397 	if (mvp->v_usecount != 0) {
398 		TAILQ_REMOVE(&mvp->v_mount->mnt_vnodelist, mvip, vi_mntvnodes);
399 		mvp->v_usecount = 0;
400 	}
401 	mutex_exit(&mntvnode_lock);
402 	vnfree_marker(mvp);
403 }
404 
405 static struct vnode *
406 vfs_vnode_iterator_next1(struct vnode_iterator *vni,
407     bool (*f)(void *, struct vnode *), void *cl, bool do_wait)
408 {
409 	vnode_impl_t *mvip = &vni->vi_vnode;
410 	struct mount *mp = VIMPL_TO_VNODE(mvip)->v_mount;
411 	vnode_t *vp;
412 	vnode_impl_t *vip;
413 	int error;
414 
415 	KASSERT(vnis_marker(VIMPL_TO_VNODE(mvip)));
416 
417 	do {
418 		mutex_enter(&mntvnode_lock);
419 		vip = TAILQ_NEXT(mvip, vi_mntvnodes);
420 		TAILQ_REMOVE(&mp->mnt_vnodelist, mvip, vi_mntvnodes);
421 		VIMPL_TO_VNODE(mvip)->v_usecount = 0;
422 again:
423 		vp = VIMPL_TO_VNODE(vip);
424 		if (vp == NULL) {
425 	       		mutex_exit(&mntvnode_lock);
426 	       		return NULL;
427 		}
428 		mutex_enter(vp->v_interlock);
429 		if (vnis_marker(vp) ||
430 		    vdead_check(vp, (do_wait ? 0 : VDEAD_NOWAIT)) ||
431 		    (f && !(*f)(cl, vp))) {
432 			mutex_exit(vp->v_interlock);
433 			vip = TAILQ_NEXT(vip, vi_mntvnodes);
434 			goto again;
435 		}
436 
437 		TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vip, mvip, vi_mntvnodes);
438 		VIMPL_TO_VNODE(mvip)->v_usecount = 1;
439 		mutex_exit(&mntvnode_lock);
440 		error = vcache_vget(vp);
441 		KASSERT(error == 0 || error == ENOENT);
442 	} while (error != 0);
443 
444 	return vp;
445 }
446 
447 struct vnode *
448 vfs_vnode_iterator_next(struct vnode_iterator *vni,
449     bool (*f)(void *, struct vnode *), void *cl)
450 {
451 
452 	return vfs_vnode_iterator_next1(vni, f, cl, false);
453 }
454 
455 /*
456  * Move a vnode from one mount queue to another.
457  */
458 void
459 vfs_insmntque(vnode_t *vp, struct mount *mp)
460 {
461 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
462 	struct mount *omp;
463 
464 	KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 ||
465 	    vp->v_tag == VT_VFS);
466 
467 	mutex_enter(&mntvnode_lock);
468 	/*
469 	 * Delete from old mount point vnode list, if on one.
470 	 */
471 	if ((omp = vp->v_mount) != NULL)
472 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vip, vi_mntvnodes);
473 	/*
474 	 * Insert into list of vnodes for the new mount point, if
475 	 * available.  The caller must take a reference on the mount
476 	 * structure and donate to the vnode.
477 	 */
478 	if ((vp->v_mount = mp) != NULL)
479 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vip, vi_mntvnodes);
480 	mutex_exit(&mntvnode_lock);
481 
482 	if (omp != NULL) {
483 		/* Release reference to old mount. */
484 		vfs_rele(omp);
485 	}
486 }
487 
488 /*
489  * Remove any vnodes in the vnode table belonging to mount point mp.
490  *
491  * If FORCECLOSE is not specified, there should not be any active ones,
492  * return error if any are found (nb: this is a user error, not a
493  * system error). If FORCECLOSE is specified, detach any active vnodes
494  * that are found.
495  *
496  * If WRITECLOSE is set, only flush out regular file vnodes open for
497  * writing.
498  *
499  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
500  */
501 #ifdef DEBUG
502 int busyprt = 0;	/* print out busy vnodes */
503 struct ctldebug debug1 = { "busyprt", &busyprt };
504 #endif
505 
506 static vnode_t *
507 vflushnext(struct vnode_iterator *marker, int *when)
508 {
509 	if (hardclock_ticks > *when) {
510 		yield();
511 		*when = hardclock_ticks + hz / 10;
512 	}
513 	return vfs_vnode_iterator_next1(marker, NULL, NULL, true);
514 }
515 
516 /*
517  * Flush one vnode.  Referenced on entry, unreferenced on return.
518  */
519 static int
520 vflush_one(vnode_t *vp, vnode_t *skipvp, int flags)
521 {
522 	int error;
523 	struct vattr vattr;
524 
525 	if (vp == skipvp ||
526 	    ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM))) {
527 		vrele(vp);
528 		return 0;
529 	}
530 	/*
531 	 * If WRITECLOSE is set, only flush out regular file
532 	 * vnodes open for writing or open and unlinked.
533 	 */
534 	if ((flags & WRITECLOSE)) {
535 		if (vp->v_type != VREG) {
536 			vrele(vp);
537 			return 0;
538 		}
539 		error = vn_lock(vp, LK_EXCLUSIVE);
540 		if (error) {
541 			KASSERT(error == ENOENT);
542 			vrele(vp);
543 			return 0;
544 		}
545 		error = VOP_FSYNC(vp, curlwp->l_cred, FSYNC_WAIT, 0, 0);
546 		if (error == 0)
547 			error = VOP_GETATTR(vp, &vattr, curlwp->l_cred);
548 		VOP_UNLOCK(vp);
549 		if (error) {
550 			vrele(vp);
551 			return error;
552 		}
553 		if (vp->v_writecount == 0 && vattr.va_nlink > 0) {
554 			vrele(vp);
555 			return 0;
556 		}
557 	}
558 	/*
559 	 * First try to recycle the vnode.
560 	 */
561 	if (vrecycle(vp))
562 		return 0;
563 	/*
564 	 * If FORCECLOSE is set, forcibly close the vnode.
565 	 * For block or character devices, revert to an
566 	 * anonymous device.  For all other files, just
567 	 * kill them.
568 	 */
569 	if (flags & FORCECLOSE) {
570 		if (vp->v_usecount > 1 &&
571 		    (vp->v_type == VBLK || vp->v_type == VCHR))
572 			vcache_make_anon(vp);
573 		else
574 			vgone(vp);
575 		return 0;
576 	}
577 	vrele(vp);
578 	return EBUSY;
579 }
580 
581 int
582 vflush(struct mount *mp, vnode_t *skipvp, int flags)
583 {
584 	vnode_t *vp;
585 	struct vnode_iterator *marker;
586 	int busy, error, when, retries = 2;
587 
588 	do {
589 		busy = error = when = 0;
590 
591 		/*
592 		 * First, flush out any vnode references from the
593 		 * deferred vrele list.
594 		 */
595 		vrele_flush(mp);
596 
597 		vfs_vnode_iterator_init(mp, &marker);
598 
599 		while ((vp = vflushnext(marker, &when)) != NULL) {
600 			error = vflush_one(vp, skipvp, flags);
601 			if (error == EBUSY) {
602 				error = 0;
603 				busy++;
604 #ifdef DEBUG
605 				if (busyprt && retries == 0)
606 					vprint("vflush: busy vnode", vp);
607 #endif
608 			} else if (error != 0) {
609 				break;
610 			}
611 		}
612 
613 		vfs_vnode_iterator_destroy(marker);
614 	} while (error == 0 && busy > 0 && retries-- > 0);
615 
616 	if (error)
617 		return error;
618 	if (busy)
619 		return EBUSY;
620 	return 0;
621 }
622 
623 /*
624  * Mount a file system.
625  */
626 
627 /*
628  * Scan all active processes to see if any of them have a current or root
629  * directory onto which the new filesystem has just been  mounted. If so,
630  * replace them with the new mount point.
631  */
632 static void
633 mount_checkdirs(vnode_t *olddp)
634 {
635 	vnode_t *newdp, *rele1, *rele2;
636 	struct cwdinfo *cwdi;
637 	struct proc *p;
638 	bool retry;
639 
640 	if (olddp->v_usecount == 1) {
641 		return;
642 	}
643 	if (VFS_ROOT(olddp->v_mountedhere, &newdp))
644 		panic("mount: lost mount");
645 
646 	do {
647 		retry = false;
648 		mutex_enter(proc_lock);
649 		PROCLIST_FOREACH(p, &allproc) {
650 			if ((cwdi = p->p_cwdi) == NULL)
651 				continue;
652 			/*
653 			 * Cannot change to the old directory any more,
654 			 * so even if we see a stale value it is not a
655 			 * problem.
656 			 */
657 			if (cwdi->cwdi_cdir != olddp &&
658 			    cwdi->cwdi_rdir != olddp)
659 				continue;
660 			retry = true;
661 			rele1 = NULL;
662 			rele2 = NULL;
663 			atomic_inc_uint(&cwdi->cwdi_refcnt);
664 			mutex_exit(proc_lock);
665 			rw_enter(&cwdi->cwdi_lock, RW_WRITER);
666 			if (cwdi->cwdi_cdir == olddp) {
667 				rele1 = cwdi->cwdi_cdir;
668 				vref(newdp);
669 				cwdi->cwdi_cdir = newdp;
670 			}
671 			if (cwdi->cwdi_rdir == olddp) {
672 				rele2 = cwdi->cwdi_rdir;
673 				vref(newdp);
674 				cwdi->cwdi_rdir = newdp;
675 			}
676 			rw_exit(&cwdi->cwdi_lock);
677 			cwdfree(cwdi);
678 			if (rele1 != NULL)
679 				vrele(rele1);
680 			if (rele2 != NULL)
681 				vrele(rele2);
682 			mutex_enter(proc_lock);
683 			break;
684 		}
685 		mutex_exit(proc_lock);
686 	} while (retry);
687 
688 	if (rootvnode == olddp) {
689 		vrele(rootvnode);
690 		vref(newdp);
691 		rootvnode = newdp;
692 	}
693 	vput(newdp);
694 }
695 
696 /*
697  * Start extended attributes
698  */
699 static int
700 start_extattr(struct mount *mp)
701 {
702 	int error;
703 
704 	error = VFS_EXTATTRCTL(mp, EXTATTR_CMD_START, NULL, 0, NULL);
705 	if (error)
706 		printf("%s: failed to start extattr: error = %d\n",
707 		       mp->mnt_stat.f_mntonname, error);
708 
709 	return error;
710 }
711 
712 int
713 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops,
714     const char *path, int flags, void *data, size_t *data_len)
715 {
716 	vnode_t *vp = *vpp;
717 	struct mount *mp;
718 	struct pathbuf *pb;
719 	struct nameidata nd;
720 	int error;
721 
722 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
723 	    KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data);
724 	if (error) {
725 		vfs_delref(vfsops);
726 		return error;
727 	}
728 
729 	/* Cannot make a non-dir a mount-point (from here anyway). */
730 	if (vp->v_type != VDIR) {
731 		vfs_delref(vfsops);
732 		return ENOTDIR;
733 	}
734 
735 	if (flags & MNT_EXPORTED) {
736 		vfs_delref(vfsops);
737 		return EINVAL;
738 	}
739 
740 	if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) {
741 		vfs_delref(vfsops);
742 		return ENOMEM;
743 	}
744 
745 	mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred);
746 
747 	/*
748 	 * The underlying file system may refuse the mount for
749 	 * various reasons.  Allow the user to force it to happen.
750 	 *
751 	 * Set the mount level flags.
752 	 */
753 	mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE);
754 
755 	mutex_enter(&mp->mnt_updating);
756 	error = VFS_MOUNT(mp, path, data, data_len);
757 	mp->mnt_flag &= ~MNT_OP_FLAGS;
758 
759 	if (error != 0)
760 		goto err_unmounted;
761 
762 	/*
763 	 * Validate and prepare the mount point.
764 	 */
765 	error = pathbuf_copyin(path, &pb);
766 	if (error != 0) {
767 		goto err_mounted;
768 	}
769 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
770 	error = namei(&nd);
771 	pathbuf_destroy(pb);
772 	if (error != 0) {
773 		goto err_mounted;
774 	}
775 	if (nd.ni_vp != vp) {
776 		vput(nd.ni_vp);
777 		error = EINVAL;
778 		goto err_mounted;
779 	}
780 	if (vp->v_mountedhere != NULL) {
781 		vput(nd.ni_vp);
782 		error = EBUSY;
783 		goto err_mounted;
784 	}
785 	error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0);
786 	if (error != 0) {
787 		vput(nd.ni_vp);
788 		goto err_mounted;
789 	}
790 
791 	/*
792 	 * Put the new filesystem on the mount list after root.
793 	 */
794 	cache_purge(vp);
795 	mp->mnt_iflag &= ~IMNT_WANTRDWR;
796 
797 	mountlist_append(mp);
798 	if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
799 		vfs_syncer_add_to_worklist(mp);
800 	vp->v_mountedhere = mp;
801 	vput(nd.ni_vp);
802 
803 	mount_checkdirs(vp);
804 	mutex_exit(&mp->mnt_updating);
805 
806 	/* Hold an additional reference to the mount across VFS_START(). */
807 	vfs_ref(mp);
808 	(void) VFS_STATVFS(mp, &mp->mnt_stat);
809 	error = VFS_START(mp, 0);
810        if (error) {
811 		vrele(vp);
812 	} else if (flags & MNT_EXTATTR) {
813 		(void)start_extattr(mp);
814 	}
815 	/* Drop reference held for VFS_START(). */
816 	vfs_rele(mp);
817 	*vpp = NULL;
818 	return error;
819 
820 err_mounted:
821 	if (VFS_UNMOUNT(mp, MNT_FORCE) != 0)
822 		panic("Unmounting fresh file system failed");
823 
824 err_unmounted:
825 	vp->v_mountedhere = NULL;
826 	mutex_exit(&mp->mnt_updating);
827 	vfs_rele(mp);
828 
829 	return error;
830 }
831 
832 /*
833  * Do the actual file system unmount.  File system is assumed to have
834  * been locked by the caller.
835  *
836  * => Caller hold reference to the mount, explicitly for dounmount().
837  */
838 int
839 dounmount(struct mount *mp, int flags, struct lwp *l)
840 {
841 	vnode_t *coveredvp;
842 	int error, async, used_syncer, used_extattr;
843 	const bool was_suspended = fstrans_is_owner(mp);
844 
845 #if NVERIEXEC > 0
846 	error = veriexec_unmountchk(mp);
847 	if (error)
848 		return (error);
849 #endif /* NVERIEXEC > 0 */
850 
851 	if (!was_suspended) {
852 		error = vfs_suspend(mp, 0);
853 		if (error) {
854 			return error;
855 		}
856 	}
857 
858 	KASSERT((mp->mnt_iflag & IMNT_GONE) == 0);
859 
860 	used_syncer = (mp->mnt_iflag & IMNT_ONWORKLIST) != 0;
861 	used_extattr = mp->mnt_flag & MNT_EXTATTR;
862 
863 	mp->mnt_iflag |= IMNT_UNMOUNT;
864 	mutex_enter(&mp->mnt_updating);
865 	async = mp->mnt_flag & MNT_ASYNC;
866 	mp->mnt_flag &= ~MNT_ASYNC;
867 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
868 	if (used_syncer)
869 		vfs_syncer_remove_from_worklist(mp);
870 	error = 0;
871 	if (((mp->mnt_flag & MNT_RDONLY) == 0) && ((flags & MNT_FORCE) == 0)) {
872 		error = VFS_SYNC(mp, MNT_WAIT, l->l_cred);
873 	}
874 	if (error == 0 || (flags & MNT_FORCE)) {
875 		error = VFS_UNMOUNT(mp, flags);
876 	}
877 	if (error) {
878 		mp->mnt_iflag &= ~IMNT_UNMOUNT;
879 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
880 			vfs_syncer_add_to_worklist(mp);
881 		mp->mnt_flag |= async;
882 		mutex_exit(&mp->mnt_updating);
883 		if (!was_suspended)
884 			vfs_resume(mp);
885 		if (used_extattr) {
886 			if (start_extattr(mp) != 0)
887 				mp->mnt_flag &= ~MNT_EXTATTR;
888 			else
889 				mp->mnt_flag |= MNT_EXTATTR;
890 		}
891 		return (error);
892 	}
893 	mutex_exit(&mp->mnt_updating);
894 
895 	/*
896 	 * mark filesystem as gone to prevent further umounts
897 	 * after mnt_umounting lock is gone, this also prevents
898 	 * vfs_busy() from succeeding.
899 	 */
900 	mp->mnt_iflag |= IMNT_GONE;
901 	if (!was_suspended)
902 		vfs_resume(mp);
903 
904 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
905 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY);
906 		coveredvp->v_mountedhere = NULL;
907 		VOP_UNLOCK(coveredvp);
908 	}
909 	mountlist_remove(mp);
910 	if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL)
911 		panic("unmount: dangling vnode");
912 	vfs_hooks_unmount(mp);
913 
914 	vfs_rele(mp);	/* reference from mount() */
915 	if (coveredvp != NULLVP) {
916 		vrele(coveredvp);
917 	}
918 	return (0);
919 }
920 
921 /*
922  * Unmount all file systems.
923  * We traverse the list in reverse order under the assumption that doing so
924  * will avoid needing to worry about dependencies.
925  */
926 bool
927 vfs_unmountall(struct lwp *l)
928 {
929 
930 	printf("unmounting file systems...\n");
931 	return vfs_unmountall1(l, true, true);
932 }
933 
934 static void
935 vfs_unmount_print(struct mount *mp, const char *pfx)
936 {
937 
938 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
939 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
940 	    mp->mnt_stat.f_fstypename);
941 }
942 
943 /*
944  * Return the mount with the highest generation less than "gen".
945  */
946 static struct mount *
947 vfs_unmount_next(uint64_t gen)
948 {
949 	mount_iterator_t *iter;
950 	struct mount *mp, *nmp;
951 
952 	nmp = NULL;
953 
954 	mountlist_iterator_init(&iter);
955 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
956 		if ((nmp == NULL || mp->mnt_gen > nmp->mnt_gen) &&
957 		    mp->mnt_gen < gen) {
958 			if (nmp != NULL)
959 				vfs_rele(nmp);
960 			nmp = mp;
961 			vfs_ref(nmp);
962 		}
963 	}
964 	mountlist_iterator_destroy(iter);
965 
966 	return nmp;
967 }
968 
969 bool
970 vfs_unmount_forceone(struct lwp *l)
971 {
972 	struct mount *mp;
973 	int error;
974 
975 	mp = vfs_unmount_next(mountgen);
976 	if (mp == NULL) {
977 		return false;
978 	}
979 
980 #ifdef DEBUG
981 	printf("forcefully unmounting %s (%s)...\n",
982 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
983 #endif
984 	if ((error = dounmount(mp, MNT_FORCE, l)) == 0) {
985 		vfs_unmount_print(mp, "forcefully ");
986 		return true;
987 	} else {
988 		vfs_rele(mp);
989 	}
990 
991 #ifdef DEBUG
992 	printf("forceful unmount of %s failed with error %d\n",
993 	    mp->mnt_stat.f_mntonname, error);
994 #endif
995 
996 	return false;
997 }
998 
999 bool
1000 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
1001 {
1002 	struct mount *mp;
1003 	bool any_error = false, progress = false;
1004 	uint64_t gen;
1005 	int error;
1006 
1007 	gen = mountgen;
1008 	for (;;) {
1009 		mp = vfs_unmount_next(gen);
1010 		if (mp == NULL)
1011 			break;
1012 		gen = mp->mnt_gen;
1013 
1014 #ifdef DEBUG
1015 		printf("unmounting %p %s (%s)...\n",
1016 		    (void *)mp, mp->mnt_stat.f_mntonname,
1017 		    mp->mnt_stat.f_mntfromname);
1018 #endif
1019 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
1020 			vfs_unmount_print(mp, "");
1021 			progress = true;
1022 		} else {
1023 			vfs_rele(mp);
1024 			if (verbose) {
1025 				printf("unmount of %s failed with error %d\n",
1026 				    mp->mnt_stat.f_mntonname, error);
1027 			}
1028 			any_error = true;
1029 		}
1030 	}
1031 	if (verbose) {
1032 		printf("unmounting done\n");
1033 	}
1034 	if (any_error && verbose) {
1035 		printf("WARNING: some file systems would not unmount\n");
1036 	}
1037 	return progress;
1038 }
1039 
1040 void
1041 vfs_sync_all(struct lwp *l)
1042 {
1043 	printf("syncing disks... ");
1044 
1045 	/* remove user processes from run queue */
1046 	suspendsched();
1047 	(void)spl0();
1048 
1049 	/* avoid coming back this way again if we panic. */
1050 	doing_shutdown = 1;
1051 
1052 	do_sys_sync(l);
1053 
1054 	/* Wait for sync to finish. */
1055 	if (buf_syncwait() != 0) {
1056 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
1057 		Debugger();
1058 #endif
1059 		printf("giving up\n");
1060 		return;
1061 	} else
1062 		printf("done\n");
1063 }
1064 
1065 /*
1066  * Sync and unmount file systems before shutting down.
1067  */
1068 void
1069 vfs_shutdown(void)
1070 {
1071 	lwp_t *l = curlwp;
1072 
1073 	vfs_sync_all(l);
1074 
1075 	/*
1076 	 * If we have paniced - do not make the situation potentially
1077 	 * worse by unmounting the file systems.
1078 	 */
1079 	if (panicstr != NULL) {
1080 		return;
1081 	}
1082 
1083 	/* Unmount file systems. */
1084 	vfs_unmountall(l);
1085 }
1086 
1087 /*
1088  * Print a list of supported file system types (used by vfs_mountroot)
1089  */
1090 static void
1091 vfs_print_fstypes(void)
1092 {
1093 	struct vfsops *v;
1094 	int cnt = 0;
1095 
1096 	mutex_enter(&vfs_list_lock);
1097 	LIST_FOREACH(v, &vfs_list, vfs_list)
1098 		++cnt;
1099 	mutex_exit(&vfs_list_lock);
1100 
1101 	if (cnt == 0) {
1102 		printf("WARNING: No file system modules have been loaded.\n");
1103 		return;
1104 	}
1105 
1106 	printf("Supported file systems:");
1107 	mutex_enter(&vfs_list_lock);
1108 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1109 		printf(" %s", v->vfs_name);
1110 	}
1111 	mutex_exit(&vfs_list_lock);
1112 	printf("\n");
1113 }
1114 
1115 /*
1116  * Mount the root file system.  If the operator didn't specify a
1117  * file system to use, try all possible file systems until one
1118  * succeeds.
1119  */
1120 int
1121 vfs_mountroot(void)
1122 {
1123 	struct vfsops *v;
1124 	int error = ENODEV;
1125 
1126 	if (root_device == NULL)
1127 		panic("vfs_mountroot: root device unknown");
1128 
1129 	switch (device_class(root_device)) {
1130 	case DV_IFNET:
1131 		if (rootdev != NODEV)
1132 			panic("vfs_mountroot: rootdev set for DV_IFNET "
1133 			    "(0x%llx -> %llu,%llu)",
1134 			    (unsigned long long)rootdev,
1135 			    (unsigned long long)major(rootdev),
1136 			    (unsigned long long)minor(rootdev));
1137 		break;
1138 
1139 	case DV_DISK:
1140 		if (rootdev == NODEV)
1141 			panic("vfs_mountroot: rootdev not set for DV_DISK");
1142 	        if (bdevvp(rootdev, &rootvp))
1143 	                panic("vfs_mountroot: can't get vnode for rootdev");
1144 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
1145 		if (error) {
1146 			printf("vfs_mountroot: can't open root device\n");
1147 			return (error);
1148 		}
1149 		break;
1150 
1151 	case DV_VIRTUAL:
1152 		break;
1153 
1154 	default:
1155 		printf("%s: inappropriate for root file system\n",
1156 		    device_xname(root_device));
1157 		return (ENODEV);
1158 	}
1159 
1160 	/*
1161 	 * If user specified a root fs type, use it.  Make sure the
1162 	 * specified type exists and has a mount_root()
1163 	 */
1164 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
1165 		v = vfs_getopsbyname(rootfstype);
1166 		error = EFTYPE;
1167 		if (v != NULL) {
1168 			if (v->vfs_mountroot != NULL) {
1169 				error = (v->vfs_mountroot)();
1170 			}
1171 			v->vfs_refcount--;
1172 		}
1173 		goto done;
1174 	}
1175 
1176 	/*
1177 	 * Try each file system currently configured into the kernel.
1178 	 */
1179 	mutex_enter(&vfs_list_lock);
1180 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1181 		if (v->vfs_mountroot == NULL)
1182 			continue;
1183 #ifdef DEBUG
1184 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
1185 #endif
1186 		v->vfs_refcount++;
1187 		mutex_exit(&vfs_list_lock);
1188 		error = (*v->vfs_mountroot)();
1189 		mutex_enter(&vfs_list_lock);
1190 		v->vfs_refcount--;
1191 		if (!error) {
1192 			aprint_normal("root file system type: %s\n",
1193 			    v->vfs_name);
1194 			break;
1195 		}
1196 	}
1197 	mutex_exit(&vfs_list_lock);
1198 
1199 	if (v == NULL) {
1200 		vfs_print_fstypes();
1201 		printf("no file system for %s", device_xname(root_device));
1202 		if (device_class(root_device) == DV_DISK)
1203 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
1204 		printf("\n");
1205 		error = EFTYPE;
1206 	}
1207 
1208 done:
1209 	if (error && device_class(root_device) == DV_DISK) {
1210 		VOP_CLOSE(rootvp, FREAD, FSCRED);
1211 		vrele(rootvp);
1212 	}
1213 	if (error == 0) {
1214 		mount_iterator_t *iter;
1215 		struct mount *mp;
1216 		extern struct cwdinfo cwdi0;
1217 
1218 		mountlist_iterator_init(&iter);
1219 		mp = mountlist_iterator_next(iter);
1220 		KASSERT(mp != NULL);
1221 		mountlist_iterator_destroy(iter);
1222 
1223 		mp->mnt_flag |= MNT_ROOTFS;
1224 		mp->mnt_op->vfs_refcount++;
1225 
1226 		/*
1227 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
1228 		 * reference it.
1229 		 */
1230 		error = VFS_ROOT(mp, &rootvnode);
1231 		if (error)
1232 			panic("cannot find root vnode, error=%d", error);
1233 		cwdi0.cwdi_cdir = rootvnode;
1234 		vref(cwdi0.cwdi_cdir);
1235 		VOP_UNLOCK(rootvnode);
1236 		cwdi0.cwdi_rdir = NULL;
1237 
1238 		/*
1239 		 * Now that root is mounted, we can fixup initproc's CWD
1240 		 * info.  All other processes are kthreads, which merely
1241 		 * share proc0's CWD info.
1242 		 */
1243 		initproc->p_cwdi->cwdi_cdir = rootvnode;
1244 		vref(initproc->p_cwdi->cwdi_cdir);
1245 		initproc->p_cwdi->cwdi_rdir = NULL;
1246 		/*
1247 		 * Enable loading of modules from the filesystem
1248 		 */
1249 		module_load_vfs_init();
1250 
1251 	}
1252 	return (error);
1253 }
1254 
1255 /*
1256  * mount_specific_key_create --
1257  *	Create a key for subsystem mount-specific data.
1258  */
1259 int
1260 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1261 {
1262 
1263 	return specificdata_key_create(mount_specificdata_domain, keyp, dtor);
1264 }
1265 
1266 /*
1267  * mount_specific_key_delete --
1268  *	Delete a key for subsystem mount-specific data.
1269  */
1270 void
1271 mount_specific_key_delete(specificdata_key_t key)
1272 {
1273 
1274 	specificdata_key_delete(mount_specificdata_domain, key);
1275 }
1276 
1277 /*
1278  * mount_initspecific --
1279  *	Initialize a mount's specificdata container.
1280  */
1281 void
1282 mount_initspecific(struct mount *mp)
1283 {
1284 	int error __diagused;
1285 
1286 	error = specificdata_init(mount_specificdata_domain,
1287 				  &mp->mnt_specdataref);
1288 	KASSERT(error == 0);
1289 }
1290 
1291 /*
1292  * mount_finispecific --
1293  *	Finalize a mount's specificdata container.
1294  */
1295 void
1296 mount_finispecific(struct mount *mp)
1297 {
1298 
1299 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
1300 }
1301 
1302 /*
1303  * mount_getspecific --
1304  *	Return mount-specific data corresponding to the specified key.
1305  */
1306 void *
1307 mount_getspecific(struct mount *mp, specificdata_key_t key)
1308 {
1309 
1310 	return specificdata_getspecific(mount_specificdata_domain,
1311 					 &mp->mnt_specdataref, key);
1312 }
1313 
1314 /*
1315  * mount_setspecific --
1316  *	Set mount-specific data corresponding to the specified key.
1317  */
1318 void
1319 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
1320 {
1321 
1322 	specificdata_setspecific(mount_specificdata_domain,
1323 				 &mp->mnt_specdataref, key, data);
1324 }
1325 
1326 /*
1327  * Check to see if a filesystem is mounted on a block device.
1328  */
1329 int
1330 vfs_mountedon(vnode_t *vp)
1331 {
1332 	vnode_t *vq;
1333 	int error = 0;
1334 
1335 	if (vp->v_type != VBLK)
1336 		return ENOTBLK;
1337 	if (spec_node_getmountedfs(vp) != NULL)
1338 		return EBUSY;
1339 	if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, &vq) == 0) {
1340 		if (spec_node_getmountedfs(vq) != NULL)
1341 			error = EBUSY;
1342 		vrele(vq);
1343 	}
1344 
1345 	return error;
1346 }
1347 
1348 /*
1349  * Check if a device pointed to by vp is mounted.
1350  *
1351  * Returns:
1352  *   EINVAL	if it's not a disk
1353  *   EBUSY	if it's a disk and mounted
1354  *   0		if it's a disk and not mounted
1355  */
1356 int
1357 rawdev_mounted(vnode_t *vp, vnode_t **bvpp)
1358 {
1359 	vnode_t *bvp;
1360 	dev_t dev;
1361 	int d_type;
1362 
1363 	bvp = NULL;
1364 	d_type = D_OTHER;
1365 
1366 	if (iskmemvp(vp))
1367 		return EINVAL;
1368 
1369 	switch (vp->v_type) {
1370 	case VCHR: {
1371 		const struct cdevsw *cdev;
1372 
1373 		dev = vp->v_rdev;
1374 		cdev = cdevsw_lookup(dev);
1375 		if (cdev != NULL) {
1376 			dev_t blkdev;
1377 
1378 			blkdev = devsw_chr2blk(dev);
1379 			if (blkdev != NODEV) {
1380 				if (vfinddev(blkdev, VBLK, &bvp) != 0) {
1381 					d_type = (cdev->d_flag & D_TYPEMASK);
1382 					/* XXX: what if bvp disappears? */
1383 					vrele(bvp);
1384 				}
1385 			}
1386 		}
1387 
1388 		break;
1389 		}
1390 
1391 	case VBLK: {
1392 		const struct bdevsw *bdev;
1393 
1394 		dev = vp->v_rdev;
1395 		bdev = bdevsw_lookup(dev);
1396 		if (bdev != NULL)
1397 			d_type = (bdev->d_flag & D_TYPEMASK);
1398 
1399 		bvp = vp;
1400 
1401 		break;
1402 		}
1403 
1404 	default:
1405 		break;
1406 	}
1407 
1408 	if (d_type != D_DISK)
1409 		return EINVAL;
1410 
1411 	if (bvpp != NULL)
1412 		*bvpp = bvp;
1413 
1414 	/*
1415 	 * XXX: This is bogus. We should be failing the request
1416 	 * XXX: not only if this specific slice is mounted, but
1417 	 * XXX: if it's on a disk with any other mounted slice.
1418 	 */
1419 	if (vfs_mountedon(bvp))
1420 		return EBUSY;
1421 
1422 	return 0;
1423 }
1424 
1425 /*
1426  * Make a 'unique' number from a mount type name.
1427  */
1428 long
1429 makefstype(const char *type)
1430 {
1431 	long rv;
1432 
1433 	for (rv = 0; *type; type++) {
1434 		rv <<= 2;
1435 		rv ^= *type;
1436 	}
1437 	return rv;
1438 }
1439 
1440 static struct mountlist_entry *
1441 mountlist_alloc(enum mountlist_type type, struct mount *mp)
1442 {
1443 	struct mountlist_entry *me;
1444 
1445 	me = kmem_zalloc(sizeof(*me), KM_SLEEP);
1446 	me->me_mount = mp;
1447 	me->me_type = type;
1448 
1449 	return me;
1450 }
1451 
1452 static void
1453 mountlist_free(struct mountlist_entry *me)
1454 {
1455 
1456 	kmem_free(me, sizeof(*me));
1457 }
1458 
1459 void
1460 mountlist_iterator_init(mount_iterator_t **mip)
1461 {
1462 	struct mountlist_entry *me;
1463 
1464 	me = mountlist_alloc(ME_MARKER, NULL);
1465 	mutex_enter(&mountlist_lock);
1466 	TAILQ_INSERT_HEAD(&mountlist, me, me_list);
1467 	mutex_exit(&mountlist_lock);
1468 	*mip = (mount_iterator_t *)me;
1469 }
1470 
1471 void
1472 mountlist_iterator_destroy(mount_iterator_t *mi)
1473 {
1474 	struct mountlist_entry *marker = &mi->mi_entry;
1475 
1476 	if (marker->me_mount != NULL)
1477 		vfs_unbusy(marker->me_mount);
1478 
1479 	mutex_enter(&mountlist_lock);
1480 	TAILQ_REMOVE(&mountlist, marker, me_list);
1481 	mutex_exit(&mountlist_lock);
1482 
1483 	mountlist_free(marker);
1484 
1485 }
1486 
1487 /*
1488  * Return the next mount or NULL for this iterator.
1489  * Mark it busy on success.
1490  */
1491 static inline struct mount *
1492 _mountlist_iterator_next(mount_iterator_t *mi, bool wait)
1493 {
1494 	struct mountlist_entry *me, *marker = &mi->mi_entry;
1495 	struct mount *mp;
1496 	int error;
1497 
1498 	if (marker->me_mount != NULL) {
1499 		vfs_unbusy(marker->me_mount);
1500 		marker->me_mount = NULL;
1501 	}
1502 
1503 	mutex_enter(&mountlist_lock);
1504 	for (;;) {
1505 		KASSERT(marker->me_type == ME_MARKER);
1506 
1507 		me = TAILQ_NEXT(marker, me_list);
1508 		if (me == NULL) {
1509 			/* End of list: keep marker and return. */
1510 			mutex_exit(&mountlist_lock);
1511 			return NULL;
1512 		}
1513 		TAILQ_REMOVE(&mountlist, marker, me_list);
1514 		TAILQ_INSERT_AFTER(&mountlist, me, marker, me_list);
1515 
1516 		/* Skip other markers. */
1517 		if (me->me_type != ME_MOUNT)
1518 			continue;
1519 
1520 		/* Take an initial reference for vfs_busy() below. */
1521 		mp = me->me_mount;
1522 		KASSERT(mp != NULL);
1523 		vfs_ref(mp);
1524 		mutex_exit(&mountlist_lock);
1525 
1526 		/* Try to mark this mount busy and return on success. */
1527 		if (wait)
1528 			error = vfs_busy(mp);
1529 		else
1530 			error = vfs_trybusy(mp);
1531 		if (error == 0) {
1532 			vfs_rele(mp);
1533 			marker->me_mount = mp;
1534 			return mp;
1535 		}
1536 		vfs_rele(mp);
1537 		mutex_enter(&mountlist_lock);
1538 	}
1539 }
1540 
1541 struct mount *
1542 mountlist_iterator_next(mount_iterator_t *mi)
1543 {
1544 
1545 	return _mountlist_iterator_next(mi, true);
1546 }
1547 
1548 struct mount *
1549 mountlist_iterator_trynext(mount_iterator_t *mi)
1550 {
1551 
1552 	return _mountlist_iterator_next(mi, false);
1553 }
1554 
1555 /*
1556  * Attach new mount to the end of the mount list.
1557  */
1558 void
1559 mountlist_append(struct mount *mp)
1560 {
1561 	struct mountlist_entry *me;
1562 
1563 	me = mountlist_alloc(ME_MOUNT, mp);
1564 	mutex_enter(&mountlist_lock);
1565 	TAILQ_INSERT_TAIL(&mountlist, me, me_list);
1566 	mutex_exit(&mountlist_lock);
1567 }
1568 
1569 /*
1570  * Remove mount from mount list.
1571  */void
1572 mountlist_remove(struct mount *mp)
1573 {
1574 	struct mountlist_entry *me;
1575 
1576 	mutex_enter(&mountlist_lock);
1577 	TAILQ_FOREACH(me, &mountlist, me_list)
1578 		if (me->me_type == ME_MOUNT && me->me_mount == mp)
1579 			break;
1580 	KASSERT(me != NULL);
1581 	TAILQ_REMOVE(&mountlist, me, me_list);
1582 	mutex_exit(&mountlist_lock);
1583 	mountlist_free(me);
1584 }
1585 
1586 /*
1587  * Unlocked variant to traverse the mountlist.
1588  * To be used from DDB only.
1589  */
1590 struct mount *
1591 _mountlist_next(struct mount *mp)
1592 {
1593 	struct mountlist_entry *me;
1594 
1595 	if (mp == NULL) {
1596 		me = TAILQ_FIRST(&mountlist);
1597 	} else {
1598 		TAILQ_FOREACH(me, &mountlist, me_list)
1599 			if (me->me_type == ME_MOUNT && me->me_mount == mp)
1600 				break;
1601 		if (me != NULL)
1602 			me = TAILQ_NEXT(me, me_list);
1603 	}
1604 
1605 	while (me != NULL && me->me_type != ME_MOUNT)
1606 		me = TAILQ_NEXT(me, me_list);
1607 
1608 	return (me ? me->me_mount : NULL);
1609 }
1610